runc dont kill container

[lengrongfu]

Reproduce steps：

• 1. create rootfs file

# create the top most bundle directory
$ mkdir /mycontainer
$ cd /mycontainer

# create the rootfs directory
$ mkdir rootfs

# export busybox via Docker into the rootfs directory
$ docker export $(docker create busybox) | tar -C rootfs -xvf -

• 2. create spec

$ cd mycontainer 
$ runc spec

• 3. update config.json

# Do not modify other
"terminal": true, => "terminal": false,

• 4. create and start container

$ runc create test
$ runc start test

• 5. kill test container

$ runc kill test # but kill failed

[fahedouch]

Hi @lengrongfu,

it depends on the init process of you container (default sh). The init process is a special process. It behaves differently then normal process when receiving unix signals.

the above command will send a SIGTERM for you container init process
runc kill test # but kill failed

So you have to check how your init process handles the SIGTERM, otherwise you send more powerful sig e.g ( runc kill test SIGKILL

I recommend you to this reference to help you debug your issue

[lengrongfu]

I am just doing a test, do we need special treatment for sh as a container process, to ensure that the container status can change to stopped after the user executes kill, and then runc delete can be executed normally.

[kolyshkin]

As described by @fahedouch above, all that runc kill does is sending a signal to the container's init process (which, in this case, is busybox shell). The default signal is SIGTERM, and how a process is handling it depends on the process.

If you do need to forcibly kill and delete a container, use runc delete -f <container-id> or runc kill <container-id> KILL followed by runc delete <container-id>.

I don't see how what you describe is a bug, thus moving to discussion

[lengrongfu]

I see, thank you.